Multiple autonomous underwater vehicles (AUVs) integrating multi-agent reinforcement learning (MARL) have made remarkable achievement and widely utilized for underwater search and rescue missions. However, to perform collaborative multi-AUV search efficiently in harsh and communication-constrained marine environments, challenging issues need to be addressed, such as cold-start problem and poor collaborative information fusion. To deal with these challenges, this paper proposes a MACS scheme which integrates the reasoning capabilities of large language models (LLMs) into the MARL framework to solve the cold-start problem and facilitate efficient collaborative information fusion. In MACS, to alleviate the cold-start problem of MARL caused by the lack of prior knowledge, we design a LEMACS algorithm, which leverages LLMs to infer the initial Target Probability Map (TPM) from search tasks and underwater terrain information to accelerate the search process. Furthermore, to address low efficient data exchange and fusion issue under unstable channel, we propose a LLM-enhanced link selection algorithm LESCL which integrates TPM information and AUV link metrics to optimize the link selection procedure to enhance multi-AUV cooperative search information fusion. To validate the effectiveness of the proposed algorithms, we conduct extensive numerical simulations using open-source regional underwater terrain data, such as coral reef map dataset of Arizona State University (ASU) and the terrain data of the Dongsha Islands, and the simulation results indicate that MACS achieves a search success rate of up to 95% in emergency multi-AUV cooperative search missions. The code is available at https://github.com/SDUST-smartocean/MACS.

In the Internet of Underwater Things (IoUT), mobile edge nodes serve as critical components for data harvesting and network relaying in expansive marine environments. However, their high-mobility nature induces severe Doppler shifts in OFDM-based underwater acoustic links, destroying subcarrier orthogonality and causing strong inter-carrier interference (ICI). Meanwhile, the operation of onboard propulsion systems generates significant non-Gaussian impulsive noise, further under-mining link reliability. This paper proposes a robust channel estimation framework termed Robust Dynamic Cluster-Sparse Bayesian Learning (RDC-SBL). By integrating the Complex Exponential Basis Expansion Model (CE-BEM) with a State Space Model (SSM), RDC-SBL utilizes Inverse-Wishart (IW) and Student-tpriors to track dynamic cluster evolution and adaptively suppress impulsive interference. To enable real-time edge intelligence on resource-constrained IoUT nodes, we further develop RDC-Mamba, a lightweight network distilled from RDC-SBL. By exploiting the structural isomorphism between the Mamba architecture and physical SSMs, RDC-Mamba approximates complex Bayesian posterior inference with linear-time complexity. Simulation results demonstrate that RDC-SBL achieves a 3 dB NMSE gain and reduces the BER to 8 × 10−4under strong interference. Crucially, RDC-Mamba slashes the single-frame inference time from 1.52 s to 12 ms, effectively overcoming the computational bottleneck of onboard processors to satisfy the strict low-latency requirements of dynamic IoUT networks. Sea trials in the South China Sea and the Yellow Sea further validate the framework’s effectiveness for practical underwater engineering applications.

For Underwater Internet of Things (UIoT) applications, the accurate and efficient estimation of the direction of arrival (DOA) is fundamental to technologies such as node localization and autonomous underwater vehicle (AUV) node cooperative communication. However, the low signal-to-noise ratio (SNR) and limited energy in underwater environments pose severe challenges to DOA estimation. Furthermore, existing methods typically require a large number of snapshots. To address these issues, this paper proposes the use of an adaptive wavelet denoising model to enhance the quality of underwater acoustic signals. Subsequently, a dual-branch space-time state space network (ST-SSNet) is proposed. This network consists of a time feature extraction branch (TFEB) and a space feature extraction branch (SFEB). The time branch incorporates gating units and time mixing functions into the state space model (SSM) within the Mamba framework to extract temporal features. The spatial branch uses one-dimensional convolutions in different directions and the convolutional block attention module (CBAM) to extract spatial features. Extensive simulation and sea trial experiments demonstrate that ST-SSNet outperforms other deep learning methods in various scenarios, while having lower computational complexity than other methods. Compared to ResNet18, the accuracy improves by 2.14%, and RMSE is reduced by 43.4%.

Underwater Wireless Sensor Networks (UWSNs) have been shown to overcome the environmental extremes and energy dependency issues faced by traditional IoT in marine environments, leading to rapid development in fields such as environmental monitoring and disaster warning. Among these, Autonomous Underwater Vehicles (AUVs) play a pivotal role in UWSNs. However, the mobility of AUVs poses significant challenges in detecting and controlling infected nodes due to the randomization of malicious program cross-platform infection and propagation paths. Accordingly, a mathematical model centered on the framework of epidemic theory has been proposed to study the propagation patterns of malicious programs in two coupled networks (AUVs and UWSNs). This model utilizes the mutual infection coefficient between AUVs and UWSNs to represent the cross-infection of malicious programs. In order to investigate the impact of AUV mobility on malware propagation, an improved cellular automaton model is proposed. This model combines the state transitions of epidemic theory with the cellular automaton model to represent the spatio-temporal propagation of malware. Furthermore, to attain optimal decision-making under resource constraints, we propose an optimization problem combining a mathematical model with defense strategies and use the Sand Dune Cat Swarm Optimization Algorithm (SCSO) to obtain the optimal control strategy. Finally, simulation experiments demonstrate that AUV movement and expanded communication radii exacerbate malware propagation, while also validating the influence of the basic reproduction number (R0) on malware propagation and the inhibitory effect of optimal control strategies on malware propagation.

The direct cross-boundary communication between Unmanned Aerial Vehicles (UAVs) and Autonomous Underwater Vehicles (AUVs) is a pivotal component in establishing the 6G integrated air-sea-space network, holding significant importance for applications such as marine data collection and maritime collaborative search and rescue. Nevertheless, existing solutions exhibit pronounced deficiencies in path planning efficiency, the coverage range of wireless optical communication, and edge computing load balancing, which result in a high Age of Information (AoI), severely compromising the performance of time-sensitive maritime missions. To address these challenges, this paper proposes a maritime data collection scheme called MECOS, which includes LMAR2P algorithm for UAVs path planning and MAPBal algorithm for UAVs to deal with the load balancing issue. In LMAR2P, a multi-agent deep reinforcement learning (MARL) architecture is adopted, in which we leverage the global understanding capability of large language models (LLMs) to provide state representation for MARL, in order to improve path planning efficiency. Furthermore, to solve the unbalanced computational load problem, we design a kullback-leibler (KL) divergence-based reward correction mechanism and propose a distributed adaptive offloading balancing algorithm MAPBal, which enables resource-aware task allocation to ensure load balancing and reduce data processing latency. The simulation results indicate that the MECOS scheme reduces the AoI by 23.6% and 26.4% during the data collection and data processing phases of maritime missions, respectively. This research provides a viable technical solution for practical applications such as maritime monitoring, demonstrating significant scientific value and promising application prospects. The code is available at https://github.com/SDUST-smartocean/MECOS.

The widespread deployment of neural networks has raised significant concerns regarding the integrity and privacy of model predictions, especially in malicious environments. Current approaches have explored zero-knowledge proofs for integrity verification. However, they suffer from inefficiency in proving runtime and a lack of rigorous integrity verification for non linear operations. To address these issues, we present a trustwor thy framework for Enhancing Integrity Verification of Convolutional Neural Network predictions (EIV-CNN) in a malicious model, whose key contributions are an efficient optimized sum check protocol and a robust enhanced verification mechanism. Specifically, we first propose an algorithm that enables efficient proving of both batch and collaborative CNN predictions by com bining sumcheck claims of multiple matrix multiplications into one. Moreover, we introduce a non-interactive sumcheck protocol with malicious security (NM-Sumcheck) to serve as a building block for publicly verifying matrix multiplication operations. Furthermore, we introduce a verifiable method for transforming nonlinear operations into matrix operations, enabling their sub sequent evaluation with the NM-Sumcheck protocol. Our EIV CNN provides malicious security, guarantees public verifiability, and preserves model privacy. Empirical results demonstrate that our sumcheck framework achieves constant prover time, verifier time, and proof size. Compared to the state-of-the-art, it achieves up to a 128.56× reduction in prover time, along with significant reductions in communication overhead and enhanced scalability.

Secure Aggregation (SA) is a fundamental privacy-preserving technique in Federated Learning (FL) that ensures the confidentiality of local model updates while enabling global model aggregation. Previous studies have implemented SA within the FL architecture that includes a central server. However, in a Device-to-Device (D2D) based FL, decentralized SA becomes challenging due to the lack of a central server, particularly in a zero-trust network vulnerable to Byzantine attacks. To address this issue, we present a novel Byzantine-robust decentralized SA protocol (DeSA) that guarantees the integrity of model training and aggregation while protecting the privacy of model updates. Specifically, we utilize an enhanced zk-SNARK proof system to verify the local model training process. Additionally, we propose a framework that embeds multiple zero-knowledge proofs to ensure the integrity of model aggregation, while maintaining succinct proofs and fast verification. Moreover, we present a Byzantine-robust D2D aggregation protocol that can withstand malicious nodes trying to disrupt model aggregation. To protect privacy, we develop a one-time masking method that eliminates aggregated masks through a dynamic aggregation strategy. This strategy takes into account the adjacency and trust relationships among nodes in evolving network topologies. Finally, we perform a theoretical analysis and evaluate DeSA on real-world datasets. Experimental results show that the time required to verify an embedded proof is significantly reduced compared to the time of verifying multiple proofs. Additionally, its accuracy remains robust against malicious nodes.

With the increasing number of terminal devices in the Ocean of Things (OoT), it is necessary to apply the OoT mobile edge computing (MEC) paradigm to low-Earth orbit (LEO) satellites. The aim is to support the operation of compute-intensive OoT services with LEO satellite assistance. To address the proliferation of computing services in OoT, this article proposes a satellite-assisted task offloading and resource allocation (STORA) approach for OoT edge computing, which includes a generalized framework for three-layer MEC systems in space, on the surface, and underwater. First, the MEC system energy minimization problem is described as mixed integer-nonlinear programming (MINLP) and divided into two subproblems: 1) task offloading and 2) resource allocation. Second, the task offloading subproblem is modeled as a Markov decision process (MDP). The proposed adaptive deep deterministic policy gradient (A-DDPG) algorithm jointly optimizes the offloading policy and offloading volume. In A-DDPG, a soft network update method with an adaptive updating coefficient ensures stable network updates while achieving fast convergence. Finally, the resource allocation is decomposed into a joint optimization problem involving buoy and satellite computational resources, which is shown to be convex. The Lagrange multiplier method is used to optimize the buoy-satellite resource allocation problem while also balancing edge computational load across servers. The experimental results show that STORA can reduce network energy consumption by 17.8%, increase network lifetime by 24.4%, and lower network latency by 11.5%.

Federated learning is a widely used method for collaborative machine learning without sharing local data. In this approach, participants train models using their local data, and the model updates are aggregated into a global model. However, ensuring trustworthy model training is crucial because malicious participants may not use their actual local data or may not train the model as intended, which makes it challenging to guarantee the authenticity of the data and the integrity of the model training. To address these issues, we propose a trustworthy model training scheme (TMT-FL) with verifiable authenticity and integrity. Specifically, we leverage zero-knowledge succinct non-interactive argument of knowledge (zk-SNARK) based proofs to verify the integrity of the training execution. To deal with the performance bottleneck in generating zk-SNARK proofs, we use the Chinese Remainder Theorem to optimize the convolution operation, and present an improved zk-SNARK based proof generating scheme which significantly reduces the online proving time. Besides, we adopt matrix commitment along with bloom filter to ensure the authenticity and integrity of the training datasets. Extensive experimental results demonstrate that our improved zk-SNARK scheme performs nearly$3.1\times$faster than the state-of-the-art in online proving time. Moreover, we experimentally confirm the efficiency of TMT-FL under diverse datasets in terms of computational costs, storage costs, and communication overheads.

Secure Aggregation (SA), in the Federated Learning (FL) setting, enables distributed clients to collaboratively learn a shared global model while keeping their raw data and local gradients private. However, when SA is implemented in edge-intelligence-driven FL, the open and heterogeneous environments will hinder model aggregation, slow down model convergence speed, and decrease model generalization ability. To address these issues, we present a Robust and adaptive Secure Aggregation (RaSA) protocol to guarantee robustness and privacy in the presence of non-IID data, heterogeneous system, and malicious edge servers. Specifically, we first design an adaptive weights updating strategy to address the non-IID data issue by considering the impact of both gradient similarity and gradient diversity on the model aggregation. Meanwhile, we enhance privacy protection by preventing privacy leakage from both gradients and aggregation weights. Different from previous work, we address system heterogeneity in the case of malicious attacks, and the malicious behavior from edge servers can be detected by the proposed verifiable approach. Moreover, we eliminate the influence of straggling communication links and dropouts on the model convergence by combining efficient product-coded computing with repetition-based secret sharing. Finally, we perform a theoretical analysis that proves the security of RaSA. Extensive experimental results show that RaSA can ensure model convergence without affecting the generalization ability under non-IID scenarios. Moreover, the decoding efficiency of RaSA achieves 1.33× and 6.4× faster than the state-of-the-art product-coded and one-dimensional coded computing schemes.

Split Vertical Federated Learning (Split VFL) is an increasingly popular framework for collaborative machine learning on vertically partitioned data. However, it is vulnerable to various attacks, resulting in privacy leakage and robust aggregation issues. Recent works have explored the privacy protection of raw data samples and labels, neglecting malicious attacks launched by dishonest passive parties. Since they may deviate from the protocol and launch embedding poisoning attacks and free-riding attacks, it will inevitably result in model performance loss. To address this issue, we propose a Robust Privacy-preserving forward Aggregation (RoPA) protocol, which can resist embedding poisoning attacks and free-riding attacks and protect the privacy of embedding vectors. Specifically, we first present a modified Secret-shared Non-Interactive Proofs (SNIP) algorithm to guarantee the integrity verification of embedding vectors. To prevent free-riding attacks, we also give a validity verification protocol using matrix commitment. In particular, we utilize probability checking and batch verification to improve the verification efficiency of the protocol. Moreover, we adopt arithmetic secret sharing to protect data privacy. Finally, we conduct rigorous theoretical analysis to prove the security of RoPA and evaluate the performance of RoPA. The experimental results show that the proof verification overhead of RoPA is approximately 8× lower than the original SNIP, and the model accuracy is improved by ranging from 3% to 15% under the above two malicious attacks.

As a result of the development of the Internet of Underwater Things (IoUT), underwater connected devices generate a large volume of data with varying values and time sensitivity. Previous data collection strategies cannot accommodate the varying time requirements of various data types. To address the aforementioned issues, this article proposes a cooperative data collection method (MADC-DV) for multiple autonomous underwater vehicles (AUVs) based on local global deep$Q$learning (LG-DQN) and data value, which divides data into emergency and nonemergency and achieves hybrid data collection. First, the MAC protocol for communication between AUVs and clusters is designed to divide nonemergency data into high-value data and low-value data, with low-value data not needing to reply to ACK acknowledgment packets, thereby reducing the nonemergency data collection delay. Second, nonemergency data are collected cooperatively using multiple AUVs, and the LG-DQN approach is used to plan the paths for multiple AUV data collection in order to reduce the overall energy consumption of underwater wireless sensor networks (UWSNs). Finally, emergency data are collected using a multihop routing approach to assist in the collection. A routing method is proposed to compensate for the inability of AUVs to be applied to emergency data collection. The experimental results indicate that the method can improve the network life cycle by 18.7%, reduce the delay in the collection of nonemergency data by 40%, and reduce the delay in the collection of emergency data by 26.3%, thereby meeting the varying time requirements for different types of data.

Underwater wireless sensor network (UWSN) plays a vital role in the field of ocean development and exploration. Designing a routing protocol for UWSN is a great challenge due to the characteristics of short lifetime and high delay. This paper proposes a Q-learning based routing optimization algorithm for UWSN. Two reward functions are designed based on the average residual energy of network, integrating factors such as energy information, transmission delay and link success rate to better balance transmission quality and lifetime. In addition, a holding time mechanism for packet forwarding is developed according to the priority of nodes. The simulation results show that compared to DBR and QLFR algorithms, this algorithm can effectively reduce transmission delay and prolong network lifetime.

Resource allocation faces significant challenges due to the complexity of the underwater environment. To address the problem of bandwidth assignment and improve the underwater resource utilization efficiency, this article proposes a quantum-based deep Q-network resource allocation algorithm. First, this algorithm combines factors, such as signal-to-noise ratio and data amount to construct the state space, which can better disclose the interaction between learning and environment. It also designs a unique reward function, which can guide nodes to select appropriate bandwidth, thus improving the learning capability of the deep reinforcement learning model. Furthermore, this article constructs a hybrid network model based on trainable quantum circuits, which fully utilizes various quantum gate operations to process and analyze data, predict the corresponding Q-values for actions. Simulation results show that the algorithm can reduce packet loss ratio and blocking probability while improving network bandwidth utilization.

With little or no human intervention, almost every object in Internet of Things (IoT) has ability to communicate, sense, and process information to make everything connected. Noise in complex environment has a great impact on machine-to-machine (M2M) interaction in IoT. Adaptive wavelet threshold function (AWTF)-based M2M Gaussian Noise Removal Method is proposed in this article. First, a bilateral enhanced wavelet threshold function is derived based on the adjustable zeroing window. Further, threshold is used to construct a bilateral enhanced wavelet threshold function, which can eliminate the oscillations in the existing wavelet threshold function. This ensures that there is no break in wavelet coefficients during the reconstruction process and that stable wavelet decomposition and reconstruction can be achieved. The signal-to-noise ratio (SNR) of a signal is estimated based on the variance of the noise-containing signal, and the zeroing window parameters are adjusted adaptively according to the SNR value to eliminate the noisy wavelet coefficients and improve the denoising performance. In addition, when the wavelet coefficients are reconstructed, the proposed algorithm can select a suitable threshold function according to a particular threshold value, which improves the robustness of the algorithm. “Doppler” and “Bumps” standard test signals are used as interactive signals to simulate the proposed algorithms. For “Doppler” signals, the SNR, root mean square error (RMSE), and noise suppression ratio (NSR) of AWTF increased by 8.48%, 42.59%, and 1.69%, respectively, compared with the GDES+ABC algorithm. For “Bumps” signal, the SNR, RMSE, and NSR of AWTF are improved by 11.67%, 24.46%, and 2.99%, respectively, compared with GDES+ABC algorithm. In addition, we also use different modulated signals to carry out real field experiments in Qingdao cruise ship home port, which proves the effectiveness of the proposed algorithm.

Automatic modulation recognition (AMR) technology enables receivers to automatically recognize the modulation type of the received signal for correct demodulation of the received data, but there are still many shortcomings to be addressed. To achieve accurate and efficient AMR, this paper proposes a data augmentation method for AMR, which can increase the amount of data by seven times and solve the problem of a small sample size more effectively than the existing methods. In addition, this paper proposes a concept of microscale, rationalizes the underwater acoustic signal into time series, and proposes a temporal feature extractor named One2Three block, which can extract temporal features of signals from three microscales. Finally, a spatial feature extractor named the Dual-Stream squeeze-and-excitation (SE) block is designed to abstract and synthesize more advanced spatial features for AMR. The recognition accuracy of the proposed method is verified with eight commonly used modulation modes in underwater acoustic communications on the datasets collected in the South China Sea and the Yellow Sea. The results show that the proposed method can achieve a recognition accuracy of 99% with a lower time and space complexity, and has high robustness to noisy data.

The automatic modulation recognition (AMR) enables the receiver to automatically recognize the modulation type of the received signal for achieving correct demodulation. However, the noise and interference in the underwater acoustic channel greatly influence the features extracted from the signals. In this letter, we proposed the optimizing autoencoder (OAE) and the evaluation enhanced K-nearest neighbors (EEKNN) algorithms. The combination of OAE and EEKNN not only improves the feature discrimination but also avoids the misjudgment caused by abnormal samples and realizes accurate and efficient AMR. The experimental results of the data measured in the South China sea show that the proposed method successfully recognizes eight modulation types. The recognition accuracy is up to 99.25%, and the recognition time is only 3.48 ms.

It is a huge challenge for the receiver to correctly identify the modulation types due to the complex underwater channel environment and severe noise interference. Additionally, the real-time communications have strict requirements in terms of time. In order to solve this well-known issue, in this work, we combine the automatic feature extraction and learning ability of recurrent neural network (RNN) and convolutional neural network (CNN) for designing a modulation recognition model for underwater acoustic signals. The proposed model is based on deep hybrid neural networks called recurrent and convolutional neural network (R&CNN). As compared with the traditional modulation recognition techniques, this method achieves higher recognition accuracy without manual feature extraction. The experimental results show that the validation accuracy of the proposed R&CNN’s on the Trestle data set is 98.21%. Similarly, the validation accuracy of the proposed R&CNN’s on the South China Sea data set is 99.38%. The average recognition time is 7.164ms. As compared with the conventional deep learning methods, the proposed R&CNN not only has a higher recognition accuracy, but also greatly reduces the recognition time.

With the increasingly mature sensor technology and the increasing popularity of broadband network, “the Internet-of-Everything” era is coming, and the mobile Internet of Things (IoT) is booming around the world. However, the mobile IoT communication networks face serious challenges, which are caused by the complex and variable communication environments. The mobile IoT applications can produce large-scale data, which will consume substantial energy. The transmit antenna selection (TAS) and cooperative communication schemes are commonly used to reduce the complexity and the energy consumption, which directly impact the performance of mobile IoT networks. To evaluate the performance of mobile IoT networks, it is important to analyze outage probability (OP) performance. In this article, we investigate the OP performance analysis of mobile IoT communication networks and propose an OP intelligent prediction algorithm based on an improved convolutional neural network (CNN). First, the mobile OP performance is analyzed by combining the TAS and decode-and-forward cooperative schemes, and the exact OP expressions are derived. Then, an improved CNN is designed to avoid the loss of important information, which contains the input layer, three-convolution layer, one fully connected layer, and output layer. The proposed CNN-based prediction approach is compared with the radial basis function (RBF), generalized regression (GR), Elman, and extreme learning machine (ELM) methods. The simulation results validate that the proposed CNN prediction approach can achieve a better prediction effect than RBF, Elman, GR, and ELM methods. For the CNN approach, it has a 44% increase in the prediction accuracy.

Non-orthogonal multiple access (NOMA) has been developed as a key multi-access technique for 5G. However, secure transmission remains a challenge in NOMA. Especially, the user with weakest channel is most threatened by eavesdropping, due to its highest transmit power. Two schemes are proposed to generate artificial jamming at the NOMA base station (BS), aiming at disrupting the potential eavesdropping without affecting the legitimate transmission. In the first scheme, the transmit power of artificial jamming is maximized, with its received power at each receiver higher than that of other users. Thus, the jamming signal can be eliminated via successive interference cancellation before others. When the transmit power of the BS is inadequate, the transmit jamming power is maximized with the jamming signal zero-forced at each receiver. Thus, the legitimate transmission is not affected by the jamming, and the eavesdropping can be disrupted effectively. Due to the non-convexity of these two optimization problems, we first convert them to convex ones and, then, provide an iterative algorithm to solve them. Simulation results are presented to show the effectiveness of the proposed schemes in guaranteeing the security of NOMA networks.